Embed Size (px)
Citation preview
CESIS Electronic Working Paper Series
Paper No. 332
Which Types of Relatedness Matter in Regional Growth?
- industry, occupation and education
Sofia Wixe Martin Andersson
November, 2013
The Royal Institute of technology Centre of Excellence for Science and Innovation Studies (CESIS)
http://www.cesis.se
2
Which Types of Relatedness Matter in Regional Growth?
- industry, occupation and education
Sofia WixeƟ and Martin AnderssonØ
Abstract This paper provides a conceptual discussion of relatedness, which suggests a focus on
individuals as a complement to firms and industries. The empirical relevance of the main
arguments are tested by estimating the effects of related and unrelated variety in
education and occupation among employees, as well as in industries, on regional growth.
We show that for regional productivity growth, occupational and educational related
variety matter over and above industry relatedness. This supports the conceptual
discussion put forward. The potential of productive interactions between employees in a
region is greater when there is related variety in their ‘knowledge base’. We also find that
related variety in industries is positive for employment growth but negative for
productivity growth.
Keywords: Relatedness, variety, occupation, education, regional growth.
JEL: R12, R23, J24
____________________________
Ɵ Corresponding author. Centre for Entrepreneurship and Spatial Economics (CEnSE), Jönköping International
Business School, P.O.Box 1026, 551 11 Jönköping, Sweden. E-mail: [email protected].
Ø Centre for Innovation, Research and Competence in the Learning Economy (CIRCLE), Lund University.
Acknowledgements: Sofia Wixe gratefully acknowledges financial support from the Swedish Research Council
Formas (Grant 2009-1192), the European Union/European Regional Development Fund (Grant 162 888), the
Regional Development Council of Jönköping County and the County Administrative Board in Jönköping. Martin
Andersson acknowledges financial support from FORMAS (Dnr 2011-80), as well as the Swedish Research
Council (Linnaeus Grant No. 349200680) and the Swedish Governmental Agency for Innovation Systems,
VINNOVA (Grant agreement 2010-07370).
3
1. INTRODUCTION
Within the fields of urban economics, economic geography and regional science, there has been a
long-standing debate on the effects of agglomeration economies on growth. Much work in this strand
of research focuses on the question whether industry specialization or diversity is more important in
promoting growth (Boschma and Iammarino 2009). Glaeser et al. (1992) and Henderson et al. (1995)
led the way and many researchers have followed in similar tracks. However, the point of departure for
the present paper is Frenken et al. (2005; 2007) who took the question about regional diversity, or
variety, one step further. Following Noteboom (2000), they argue that for knowledge spillovers to
enhance growth, there needs to be some sort of cognitive proximity or complementarity between
firms. A distinction was thus made between related and unrelated variety where related variety is
defined as within-industry diversity and unrelated variety as between-industry diversity. Frenken et al.
(2007) is in this regard a seminal study as it is one of the first to provide systematic evidence of that it
is not variety in general, but variety in related industries that promotes regional employment growth.
This finding has been confirmed in several studies using data from different countries and time periods
(Boschma and Iammarino 2009; Boschma et al. 2012; Hartog et al. 2012).
A conceptual issue in relation to this concerns defining relatedness. In the present paper we address
this question and argue that measuring relatedness is not straightforward since relatedness may have
many dimensions. We put forth arguments and provide empirical support for that relatedness framed
at the level of individuals, e.g. in terms of educational background and occupation, is at least as
important as relatedness in terms of industries. Indeed, the downsides of applying standard industrial
classifications to approximate relatedness have been discussed critically in a number of papers, such as
Ejermo (2005), Bishop and Gripaios (2010), Brachert et al. (2011), Desrochers and Leppälä (2011)
and Boschma et al. (2012).
In contemporary economies cities or regions tend to specialize in functions rather than in industries.
Certain occupations are found in specific cities, due to headquarters and business services being
localized in larger cities while actual production takes place in more rural areas (Duranton and Puga
2005). Larger cities are thus specialized in knowledge-intensive occupations over a wide range of
industries, which implies that variety in occupations have the potential to be at least as important for
growth as variety in industries. Since certain occupations are found in different cities it is also likely
that certain education types are found in those cities. Education and occupation are linked together,
more so than education and industry. In many cases, higher education is usually undertaken to work
within a certain range of occupations, not to work in a specific industry. The relationship between
education and occupation is still not clear-cut, with the consequence that education and occupation are
measures of quite different things. Education measures the formal, theoretical background of
4
employees while occupation is a measure of what the employees actually do in their daily work.
Indeed, occupation is commonly used as a proxy for the skills and abilities of the employees beyond
their formal education (cf. Autor et al. (2003) and Bacolod et al. (2009)). However, as Frenken et al.
(2007), most research still focus on the effect of industrial specialization (or variety) on growth.
Duranton and Puga (2004) distinguish between three types of mechanisms behind agglomeration
economies; sharing of e.g. fixed costs and risk, matching on the labor market, and learning due to
knowledge spillovers and human capital accumulation. They also emphasize that heterogeneity of
workers and firms is the foundation for these effects to materialize. From this perspective regional
variety has the potential to give rise to agglomeration economies, which may stimulate innovation and
growth. However, in a strict sense, both the matching and learning argument emphasize individuals
rather than firms. Knowledge and information may not spill over between firms per se, but rather
between employees, who channel that knowledge, in different firms. Cohen and Levinthal (1990)
discuss this in terms of firms’ absorptive capacity. There are thus arguments in favor of framing issues
of cognitive proximity and relatedness in terms of individuals. On these grounds, relatedness in terms
of employee education and occupation are emphasized in the present paper. Previous research have
indeed expanded the measures of related variety beyond industrial classifications (cf. Boschma et al.
(2012)), but the main focus is still on industry- and product level, not the specific characteristics of the
employees.
To test the empirical relevance of these ideas, we make use of data on Swedish regions and estimate
the respective influence of the different dimensions of relatedness on regional growth over a five-year
period (2002-2007). We follow the original work by Frenken et al. (2007) and compute related and
unrelated variety in terms of industries, but also include additional measures of related and unrelated
variety in terms of occupation and education of the workers in each region. The inclusion of the two
individual-based measures of occupational and educational variety constitutes the main novelty in the
empirical analysis. We further include variables reflecting general agglomeration economies as well as
a selection of control variables. The model is estimated for regions as a whole as well as for
manufacturing and service industries, respectively. This is motivated by the results from Bishop and
Gripaios (2010), which show different effects of unrelated and related variety on employment growth
in different industries.
The results show that the effects of related and unrelated variety differ considerably, both across the
different dimensions of relatedness and across sectors. This confirms the importance of expanding the
concept of relatedness beyond the industrial dimension. We find that occupational and educational
related variety matter over and above industry relatedness in explaining regional productivity growth.
Relatedness in terms of sharing a common educational background appears to be particularly
5
correlated with productivity growth in the manufacturing sector. These result may be appreciated as a
reflection of that relatedness in terms of education and occupation stimulate matching processes on
local labor markets as well as local knowledge spillovers, i.e. two established micro-foundations for
agglomeration economies. Hence, regions with a variety of jobs associated with related educations
and/or related tasks may facilitate matching externalities on the labor market as well as inter-firm
knowledge transfers through employee mobility. We also confirm the results of Frenken et al. (2007),
namely that related industry variety is negatively associated with productivity growth but positively
associated with employment growth.
The rest of the paper is organized as follows. Section 2 provides further background and motivation
for the paper, including related empirical research. Section 3 gives an overview of the data and the
variables used in the empirical application, while section 4 presents the empirical results, which give
evidence to what has been argued in the previous parts of the paper. Finally, section 5 concludes.
2. BACKGROUND AND MOTIVATION
2.1 Relatedness based on firms or individuals
Recent contributions in economic geography and regional science hold that local variety of related
industries is crucial in fostering regional growth (Frenken et al. 2007). The main argument in this
literature is that effective knowledge transfers and spillovers between activities in a region require that
they are cognitively related, though some cognitive distance is still needed to limit overlaps and to
alleviate issues of lock-in (Boschma 2005). Variety in related activities is thus maintained to stimulate
productive interactions and cross-fertilizations within a region, because it ensures cognitive proximity
while maintaining some distance (through variety). This line of reasoning builds on Noteboom’s
(2000) conjecture of “optimal cognitive distance”. Noteboom states that “information is useless if it is
not new, but it is also useless if it is so new that it cannot be understood” (Noteboom 2000, p.153).
While the general line of argument is clear, i.e. that there is a tradeoff between cognitive distance for
the sake of novelty and cognitive proximity for the sake of mutual understanding and absorptive
capacity, the question is what makes relatedness in a regional context, i.e. what characteristics of a
local economy bodes for cognitive proximity? This is an issue that is not only of academic interest, but
is also important from a policy perspective. Any discussion of policy initiatives based on the idea of
relatedness will for example bring with it a concern regarding what relatedness means and how it can
be defined and assessed.1 The idea of relatedness has had a quite large impact on both the research and
1For example, Frenken et al. (2007, p.696) conclude: “Regional policies based on supporting related variety
reduce the risk of selecting wrong activities because one takes existing regional competences as building blocks
to broaden the economic base of the region.” Picking the ‘right’ activity obviously necessitates knowledge of
what makes relatedness.
6
the policy community, and is also embedded in the European Union current regional innovation policy
concept of smart specialization (cf. McCann and Ortega-Argilés (2013).
The majority of existing analyses puts firms at the center stage and frame discussions about
relatedness in an inter-firm context. A main hypothesis is that relatedness between firms hinges on
their industry affiliations, such that firms operating in similar industries have shared competences and
thus cognitive proximity at the organizational level (cf. Boschma and Martin (2010)). In view of this,
empirical applications typically infer relatedness from pre-determined industry or product
classification schemes (Frenken et al. 2007; Boschma et al. 2012). Put simply, the potential for
productive inter-firm knowledge transfers and spillovers is assumed to be higher when firms operate in
industries that are closer to each other in the standard industrial classification system or produce
similar products (Van Oort 2013).
Following Desrochers and Leppälä (2011), it may still be argued that the firm- and industry-level
focus in the literature on relatedness downplays that much of the learning and spillovers in regions
occur at the level of individuals. That is, a strong case can be made that knowledge spillovers between
firms in a region are to a large extent attributed to spillovers between their employees. Knowledge
may flow between firms for example because their employees learn from others in their local
environment, or because employees move between them, which induces cross-fertilizations. For
instance, the theoretical as well as empirical literature on human capital externalities and local non-
market interactions (Lucas Jr. 1988; Rauch 1993; Glaeser and Scheinkman 2003) explicitly put social
interactions between individuals at center stage. Likewise, a large literature show that movements of
individuals between firms is a key source of spillovers and transfers of knowledge and information
(Almeida and Kogut 1999; Maskell and Malmberg 1999; Power and Lundmark 2004). It follows that
the central ‘agents’ in the context of spillovers and knowledge transfers are not firms per se, but
individuals. While this may seem like a trivial statement, the key point is that an emphasis on
individuals have implications for the question of the dimension of relatedness that stimulate
productive knowledge spillovers.
Accepting individuals as the main agents for knowledge spillovers suggest arguments in favor of
framing issues of cognitive proximity in terms of individual skills, experiences and knowledge. The
industry dimension can in this perspective be problematic on the grounds that experiences and
knowledge bases of individual employees have more to do with their occupational and educational
background. Many firms have for example a sharp division of labor where individual employees work
with a narrowly defined and specialized task, often matched with their university degree. Therefore,
workers’ experiences and on-the-job learning may to a large extent be considered as occupation- and
education-dependent, rather than industry-dependent. For example; a software engineer developing
7
steering-systems for industrial robots at ABB Corporation may have cognitive proximity with a
software engineer at an engineering service company, albeit the two industries are radically different
as judged by their industrial classifications. Inter-firm job switching of personnel between such firms
may also be large for the same reason. In many cases, employers also value occupational- and task-
specific experiences rather than specific industry experience when hiring new personnel.
Furthermore, there is ample evidence that industries have very different occupational and educational
composition in different regions, a phenomena dubbed ‘functional specialization’ by Duranton and
Puga (2005). This means that the experiences, and hence the position in the ‘cognitive space’, of
workers in one and the same industry may be quite different across locations. This is a potentially
important aspect that the traditional industry- and product-based measures of relatedness cannot
capture. Occupational and educational relatedness should thus better reflect cognitive proximity at an
individual level, because they depart from what employees actually do in their work, i.e. their
immediate learning and experience context, as well as their knowledge base in terms of formal
education. Our basic hypothesis is that this type of individual-level relatedness is at least as important
as industry relatedness in providing a breeding ground for productive local knowledge spillovers and
cross-fertilizations, and thus in influencing regional growth.
Similar ideas are raised by Neffke and Henning (2013) with the concept of skill relatedness. Especially
high-skilled individuals are likely to change jobs within industries that value the same types of skills.
This implies that relatedness between industries can be determined based on cross-industry labor
flows. Neffke and Henning (2013) use this in order to explain the diversification strategies of firms,
which determine regional industrial diversification. The results show that firms are more likely to
diversify into industries with which there is skill relatedness than with industries without such
relatedness or with relatedness in terms of standard industrial classifications.2 While we also
emphasize the individual-level, our contribution is different in that we study relatedness in terms of the
educational and occupational profile of regions and test their role in explaining regional growth
alongside the industry relatedness.
2 Since this characterization of relatedness is derived from actual flows of labor it is sensitive to changes in labor
flows between years and differences in labor flows between regions. In addition, if employees in certain
occupations are more likely to switch jobs than employees in other occupations there is a risk that relatedness
between industries is not properly captured by labor flows. When instead focusing on the educational
background and current occupation among employees, no matter the industry boundaries, the definition of
relatedness is time-invariant and insensitive to changes on the market.
8
2.2 Related empirical research
There are numerous empirical studies on the effects of agglomeration economies on growth. Most of
these use a broad measure of industry diversity over the economy or region as a whole, that is Jacobs
externalities are measured as unrelated rather than related variety (cf. Glaeser et al. (1992), Henderson
et al. (1995), Duranton and Puga (2000)). However, some studies acknowledge the complexity of
diversity, most notably the one by Frenken et al. (2007), who analyze the effects of unrelated and
related industry variety on growth in employment, unemployment and productivity in Dutch regions.
The results of the study show that, as expected, related variety enhances employment growth while
unrelated variety is negatively related to unemployment growth. The productivity growth of a region is
negatively associated with related variety.
Bishop and Gripaios (2010) conduct a similar study of British regions when analyzing the effects of
unrelated and related variety on employment growth in different industries. The results show that the
effects of unrelated and related variety differ across the examined industries, of which nearly 50 per
cent benefit from either one of the two forms of variety. Unrelated variety is affecting employment
growth in a larger set of industries than related variety. Regarding regional employment growth in
Germany Brachert et al. (2011) find no effects from neither unrelated nor related variety. However,
related variety in combination with functional specialization, in terms of more “white collar” workers,
positively influences employment growth. In addition, unrelated variety among “white collar” workers
and “blue collar” workers, but not R&D workers, is found to enhance growth. When analyzing growth
in Spanish provinces Boschma et al. (2012) find that related variety has a positive effect on value
added growth while unrelated variety has no growth effect. The results also show that the positive
effect is stronger when using indicators for related variety not only based on product classifications.
Boschma et al. (2012) construct one measure based on Porter’s (2003) cluster classification of
industries and one based on export data. Hartog et al. (2012) analyze the effects of related variety on
employment growth in Finnish regions. The results show that related variety in general has no effect
but that related variety in high-tech industries positively influences employment growth. No growth
effect is found from unrelated variety.
The concept of related and unrelated variety has been quite commonly applied in relation to
international trade. Saviotti and Frenken (2008) analyze the relationship between export variety and
economic development in 20 OECD countries. An increase in the growth in related export variety is
found to promote growth in GDP per capita, while an increase in the growth in unrelated export
variety has a negative effect. However, past growth in unrelated export variety positively influences
economic growth. Also Boschma and Iammarino (2009) consider export variety, when analyzing the
effects on regional economic growth in Italy. Related, but not unrelated, export variety is found to
9
significantly enhance value-added growth. In addition, as mentioned above, one of Boschma et al.’s
(2012) alternative measures of related variety is based on export data.
3. DATA AND VARIABLES
The regions referred to in the empirical part of the paper are the 290 municipalities in Sweden.3 These
are the smallest administrative units in Sweden. All variables described below are measured at the
municipal level.
3.1 Dependent variables
We employ two different dependent variables; (i) employment growth and (ii) productivity growth.
Frenken et al. (2007) find that related variety has a negative influence on productivity growth but a
positive influence on employment growth. They maintain that this is consistent with the idea that
employment growth better reflect radical innovation as such innovations are assumed to lead to the
creation of new markets and employment. Productivity growth is instead associated with process
innovations and growing capital intensity in the later stages of the product life cycle (cf. Van Oort
(2013)). That related variety spurs employment growth and not productivity growth is thus taken as
evidence that related variety has more to do with knowledge spillovers and (radical) innovation. While
this argument is conceptually appealing, it can also be argued that productivity is after all the main
determinant for long-run growth (Easterly and Levine 2002). A vast literature also document the
fundamental role played by innovation and new technology in stimulating productivity (Lööf and
Heshmati 2002; Hall and Mairesse 2006).4 Hence, there are also arguments in favor of that the
productivity growth of a region is related to innovation. A defining characteristic of urban regions, that
typically are more diversified, is indeed also higher productivity levels, not least in Sweden (cf.
Andersson and Lööf (2011)).
We measure productivity growth as the ratio between average labor productivity in 2007 and 2002.
Employment growth is measured as the ratio between total number of employed for the same set of
years. However, a region might exhibit a higher growth rate simply because it has a relatively large
share of one or more fast growing industries. As robustness test, we therefore also apply a shift-share
procedure for both productivity growth and employment growth. This implies that the growth in each
2-digit industry is weighted by the industry’s national share of production in the case of productivity
3 Gotland is excluded due to having no neighbors. Hence, the number of observations in the estimations is 289.
4In fact, it has also been argued that productivity growth may be used as an innovation indicator (Hall 2011).
10
growth and by the industry’s share of employment in the case of employment growth. This imposes
the same industrial structure in all regions and produces industry-adjusted growth rates.5
Productivity growth and employment growth are constructed for the whole private sector of the
regional economy, as well as for the manufacturing sector and the service sector separately. The
development in the public sector is more difficult to relate to spatial characteristics since it is largely
dependent on political decisions. Also agriculture, fishery and mining are excluded since these
industries are more or less spatially bounded due to immobile resources. This leaves the industries in
standard industrial classification (SIC) codes 15 to 74, of which 15 to 45 belong to the manufacturing
sector6 while 50 to 74 are service industries.
3.2 Independent variables
All independent variables are measured for the year 2002, unless otherwise specified. The six
independent variables of main interest are; related variety in industries, education and occupation, and
unrelated variety in industries, education and occupation. The entropy (or the Shannon index)
approach is commonly applied for measuring variety, see for example Jacquemin and Berry (1979),
Attaran (1986) and Frenken et al. (2007). The entropy measure has desirable properties in that it takes
the relative abundance of groups into account, and not only the absolute presence of them. The entropy
for unrelated variety measures between diversity while the entropy for related variety measures within
diversity, in e.g. industries. All entropies are calculated using employment in each group. The data is
limited to employed individuals between 20 and 64 years of age with a positive income.
For industries the 2-digit and the 5-digit SIC codes are used where each 5-digit industry belongs to a
specific 2-digit industry7. Following Attaran (1986), let Sg denote the 2-digit sets where g = 1,…, G. Eg
denotes the share of employees working in the 2-digit industry g, where Eg is measured as the share of
total regional employment. Furthermore, let Eig denote the share of employees working in the 5-digit
industry i, where i = 1,…, I, where Eig is measured as the share of employment in the respective 2-digit
industry g.
Unrelated variety in industries measures the distribution of employees between 2-digit industries.
Using the entropy approach, unrelated variety (UV) is calculated as follows:
5 See Table A1 in Van Stel and Storey (2004) for an illustration of the shift-share procedure.
6 Including construction.
7 An example: Industry 15111 and 15120 are sub industries to industry 15.
11
∑
(1)
The range of UV is from 0 to where zero variety is reached when all employees are working in
the same 2-digit industry, that is when one Eg = 1 while the rest are zero. Maximum variety, i.e. ,
is reached when there is an equal distribution of employees over all 2-digit industries, that is all Eg are
identical. (Attaran 1986)
In the same manner, the distribution of employees between 5-digit industries within each 2-digit
industry is calculated as follows:
∑
(2)
The interpretation of Equation 2 is the same as for Equation 1 with the difference that variety is
measured within each 2-digit industry rather than between the 2-digit industries. Hence, there is zero
within variety when all employees in the 2-digit industry g are working in the same 5-digit industry i,
where . Accordingly, maximum variety for industry g, , is achieved when there is an equal
distribution of employees over all 5-digit industries i, where .
The information about the degree of within variety for each 2-digit industry g, i.e. Hg, is weighted by
the relative size of industry g. Summing over all g gives the entropy measure for related variety in
industries (RV), regarding the region as a whole. These two steps are formally shown by Equation 3.
∑
(3)
Increases in the values obtained by Equation 1 and 3 imply increases in unrelated and related variety,
respectively.
Unrelated and related variety for the educational and the occupational dimension are calculated as
above, with the difference that the educational and the occupational codes are used instead of the SIC
codes. When constructing the measures for educational variety we use a combination of education
length and specialization. Employees are first categorized as either having three or more years of
higher education or not. After this categorization education specialization is used at the 2- and 4-digit
levels. This implies that employees belonging to the same 2-digit educational code and have three or
12
more years of higher education are seen as related. Education focus is divided in 26 different 2-digit
levels (see Table A1), the maximum possible G for the educational dimension is hence 52. Regarding
the occupational dimension occupational codes at the 1- and 3-digit levels are used instead of the
educational codes. Occupations are divided in 12 different 1-digit levels (see Table A2), which implies
that the maximum possible G for the occupational dimension is 12. Figures A1-3 in Appendix 2
provide maps of Sweden, showing the distribution of related and unrelated variety over municipalities
in 2002. These can also be compared to Figure A4, which presents the population density in Sweden.
Besides unrelated and related variety, we also control for the presence of general urbanization
economies measured as population density, Porter externalities or competition measured as firms per
employee, and localization economies or specialization. There are various approaches to measure
specialization, both in absolute and relative terms. In the context of localization economies absolute
specialization is relevant since it is the absolute agglomeration of employees belonging to the same
industry that has the potential to give rise to knowledge spillovers. Since an increase in unrelated
variety in industries implies a decrease in absolute industrial specialization the entropy for unrelated
variety in industries is used also as a proxy for industrial specialization (as in Aiginger and Davies
(2004)). The entropy measure is not as commonly applied to measure specialization as for example the
Herfindahl index but the two are strongly correlated (Palan 2010).
Considering previous studies of regional productivity growth, the capital-labor ratio is introduced as a
control variable when estimating growth in productivity. Size effects are controlled for by introducing
absolute values for 2002. All independent variables, besides population density, are calculated for
industry 15-74 as a whole but also for the manufacturing sector and the service sector separately.
Appendix 3 provides tables with descriptive statistics for the variables regarding the private sector as a
whole, as well as for the manufacturing sector and the service sector separately. To reduce
heteroscedasticity, as well as to facilitate the interpretation of the coefficients to be estimated, all
variables are log transformed. Appendix 4 presents correlation matrices over all independent variables
for the total private sector, the manufacturing sector and the service sector. Regarding the service
sector, there is strong correlation between related and unrelated variety in industries. To avoid
problems with multicollinearity, unrelated variety in industries is excluded when estimating models
for the service sector. In addition, the initial employment level is excluded due to problems with
multicollinearity. Besides this correlation between explanatory variables is not an issue, which is
confirmed when running post diagnostic tests for multicollinearity.
13
4. MODEL AND EMPIRICAL ESTIMATIONS
The model used as a point of departure for estimations is given by equation 5:
+ (5)
in which Growthr refers to growth in either productivity or employment in municipality r. RVr is a row
vector of related variety in industries, education and occupation, respectively and UVr is the
corresponding vector for unrelated variety. Controlr contains the set of control variables, which
depends on the Growth variable in question. εr is the error term, which by assumption is uncorrelated
with the individual variables. The parameters are estimated by ordinary least squares (OLS), with
robust standard errors in the cases where heteroscedasticity is detected. However, since the data is
based on geographic units there is a potential issue of spatial dependence, municipalities might be
correlated simply due to geographic proximity. This is tested by Moran’s I on the residuals from the
OLS regressions. In the presence of spatial autocorrelation the coefficients from the OLS regressions
are inefficient why spatial error and spatial lag models are estimated by maximum likelihood (ML).
The use of spatial econometrics to deal with spatial dependence is debated, mostly due to issues of
identification (Gibbons and Overman 2012). Hence, the spatial models are estimated mainly as
robustness checks for the OLS models and will not be analyzed in detail. More information about the
spatial models as well as the results from these estimations are presented in Appendix 5.
Table 1 presents the results for productivity growth and employment growth for the private sector as a
whole, including both unadjusted and industry-adjusted growth rates. Spatial autocorrelation is present
in the model for industry-adjusted employment growth since Moran’s I is statistically significant for
this model.
The results show that the relationships between the different dimensions of related and unrelated
variety and growth differ considerably, which gives evidence to the importance of distinguishing
between various forms of variety. As in Frenken et al. (2007), related variety in industries is positive
for employment growth, both unadjusted and adjusted, while it is negative for unadjusted productivity
growth. The results for related variety in education are consistent when it comes to growth in
productivity. Higher related variety in education is associated with higher unadjusted as well as
adjusted productivity growth.8 The estimated relationship is significantly larger than the relationship
between related variety in industries and employment growth.
8 The same patterns are found when conducting the analysis at the level of functional regions. However, due to
few observations at this level as well as recent research showing that agglomeration economies, in particular
14
Table 1. Estimated coefficients for productivity growth and employment growth in the total private
sector, standard errors in brackets.
Variables 1a
Productivity
1b
Productivity
Adjusted
2a
Employment
2b
Employment
Adjusted
RV Industry -0.130**
(0.054)
0.026
(0.075)
0.097***
(0.029)
0.162***
(0.051)
UV Industry 0.097
(0.115)
0.497***
(0.181)
-0.042
(0.082)
0.041
(0.143)
RV Education 0.562***
(0.129)
0.515***
(0.191)
0.018
(0.097)
-0.036
(0.219)
UV Education 0.227
(0.156)
0.308
(0.284)
0.023
(0.096)
-0.320*
(0.187)
RV Occupation 0.097
(0.178)
0.108
(0.199)
-0.083
(0.076)
0.032
(0.142)
UV Occupation -0.468**
(0.218)
-0.574
(0.413)
0.068
(0.150)
0.371
(0.305)
Competition 0.020
(0.030)
-0.047
(0.058)
0.049**
(0.022)
-0.097*
(0.050)
Population density 0.002
(0.008)
0.027***
(0.009)
-0.002
(0.004)
0.024***
(0.006)
Capital-labor growth 0.149***
(0.031)
0.046
(0.043)
Productivity 2002 -0.223***
(0.056)
-0.119
(0.088)
Constant 1.44***
(0.356)
0.052
(0.561)
0.143
(0.114)
-0.350
(0.243)
F-value 11.7*** 11.9*** 7.43*** 9.85***
R2 0.286 0.178 0.175 0.227
Moran’s I p-value 0.374 0.446 0.398 0.045
Observations 289 289 289 289
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in Model 1a, 1b and 2b.
Hence, the results for related variety in industries for the Swedish economy is in line with Frenken et
al.’s (2007) findings for the Netherlands. The higher employment growth in regions with greater
related variety is consistent with the argument that related variety in sectors spur knowledge spillovers
that in turn influence employment growth. The finding that related but not unrelated variety in
education is positively associated with unadjusted and adjusted productivity growth may be interpreted
as a reflection of that there needs to be complementarity in educational background among employees
for knowledge spillovers to be productive. A mechanical engineer may learn more from a materials
engineer than a sales representative, no matter their industry. One reason for this could be that they
share a common knowledge base. We find no clear association between related and unrelated variety
in occupation and the two measures of growth.
concerning knowledge spillovers, attenuate sharply with distance (cf. Baldwin et al. (2008) and Andersson et al.
(2012)), municipalities are chosen as the level of analysis.
15
Tables 2 and 3 present the results for growth in productivity and growth in employment, when the
private sector is split into manufacturing and services. Regarding productivity, we find spatial
autocorrelation in the model for adjusted growth in the manufacturing sector and unadjusted growth in
the service sector. For employment growth spatial autocorrelation is detected for the service sector
only, in both unadjusted and adjusted growth rates.
Table 2. Estimated coefficients for productivity growth in the manufacturing sector and the service
sector, standard errors in brackets.
Variables 3a
Manufacturing
3b
Manufacturing
Adjusted
4a
Services
4b
Services
Adjusted
RV Industry -0.106***
(0.036)
0.087
(0.060)
-0.133***
(0.046)
-0.129*
(0.073)
UV Industry 0.043
(0.062)
0.149
(0.102)
RV Education 0.659***
(0.150)
1.36***
(0.247)
-0.045
(0.116)
0.091
(0.167)
UV Education -0.102
(0.187)
0.168
(0.308)
0.731***
(0.261)
0.820*
(0.434)
RV Occupation 0.026
(0.077)
0.083
(0.126)
0.260***
(0.090)
0.425*
(0.226)
UV Occupation -0.186
(0.138)
0.023
(0.228)
-0.382
(0.236)
-0.492
(0.687)
Competition 0.052*
(0.028)
0.062
(0.046)
-0.052
(0.035)
0.044
(0.094)
Population density -0.006
(0.007)
0.034***
(0.011)
0.011*
(0.006)
0.017*
(0.010)
Capital labor growth 0.154***
(0.022)
-0.008
(0.036)
0.064**
(0.028)
0.087**
(0.044)
Productivity 2002 -0.211***
(0.041)
-0.132**
(0.067)
-0.492***
(0.100)
-0.300
(0.221)
Constant 1.54***
(0.269)
0.207
(0.444)
2.74***
(0.546)
1.61
(1.22)
F-value 12.4*** 15.1*** 8.65*** 2.26**
R2 0.308 0.352 0.253 0.085
Moran’s I p-value 0.143 0.001 0.028 0.346
Spatial coefficient
Observations 289 289 289 289
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in Model 4a and 4b.
The results for the manufacturing sector and the service sector in Tables 2 and 3 show both similarities
and differences from the private sector as a whole, which points to the importance of distinguishing
between different parts of the economy. The negative productivity effect from related variety in
industries is present in the service sector for both unadjusted and adjusted growth rates. In addition,
this effect is found for unadjusted growth in the manufacturing sector. Hence, the result that related
variety in industries negatively associated with productivity growth is robust. The positive
employment effect from this variable is significant only for unadjusted growth in the service sector.
16
Table 3. Estimated coefficients for employment growth in the manufacturing sector and the service
sector, standard errors in brackets.
Variables 5a
Manufacturing
5b
Manufacturing
Adjusted
6a
Services
6b
Services
Adjusted
RV Industry -0.019
(0.029)
-0.019
(0.068)
0.076***
(0.029)
0.048
(0.043)
UV Industry -0.146***
(0.050)
0.042
(0.141)
RV Education 0.170
(0.118)
0.580**
(0.240)
0.130
(0.105)
0.155
(0.202)
UV Education -0.140
(0.149)
-0.446
(0.305)
-0.264*
(0.154)
-0.165
(0.204)
RV Occupation 0.109*
(0.061)
0.377***
(0.146)
0.009
(0.093)
0.109
(0.142)
UV Occupation 0.008
(0.110)
0.315
(0.232)
-0.447*
(0.235)
-0.378
(0.333)
Competition 0.068***
(0.022)
-0.003
(0.043)
0.089***
(0.032)
0.025
(0.066)
Population density -0.016***
(0.005)
0.040***
(0.011)
0.013**
(0.005)
0.015**
(0.007)
Constant 0.261
(0.106)
-0.471**
(0.215)
0.683***
(0.187)
0.401
(0.262)
F-value 7.24*** 10.63*** 5.50*** 3.77***
R2 0.171 0.245 0.123 0.080
Moran’s I p-value 0.375 0.380 0.000 0.004
Observations 289 289 289 289
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in Model 5b, 6a and 6b.
In addition, higher unrelated variety in industries is found to be negative for unadjusted employment
growth in manufacturing. Since unrelated variety in industries measures inverse industrial
specialization this implies that positive localization economies is found for employment growth and
not productivity growth.
Regarding related variety among employees, the positive productivity effect from related variety in
education is robust for the manufacturing sector. The magnitude of this effect is particularly strong for
the industry-adjusted growth. In addition, this variable is positively enhancing adjusted employment
growth in the manufacturing sector. Hence, related variety in education is an important variable in
general and for the manufacturing sector in particular. For the private sector as a whole no effect was
found from related variety in occupation. However, the results regarding the service sector show that
related variety in occupation has a significant positive effect on both unadjusted and adjusted
productivity growth. This may be interpreted as that cognitive proximity among employees is
important for productivity growth in both sectors. For manufacturing, relatedness in terms of
educational background still matters while for services it is relatedness in current occupation. Related
variety in occupation is also positively enhancing employment growth in manufacturing, implying that
cognitive proximity is important for the manufacturing sector in general terms.
17
A somewhat contradicting result is that unrelated variety in education is productivity enhancing for
services. The magnitude of the effect is robust across unadjusted and adjusted growth rates. This
shows a difference from the manufacturing sector in that the service sector benefits from broad
diversity among employees as well. On the other hand, unrelated variety in occupation has a negative
impact on unadjusted employment growth in the service sector, and related variety in education has a
positive impact when controlling for spatial autocorrelation (see Appendix 5). Hence, it is not possible
to draw general conclusions for the service sector, as above for the manufacturing sector. This could
be due to that the service sector is comprised of a more diverse set of industries than the
manufacturing sector.
Regarding the control variables, competition has a positive impact on unadjusted employment growth,
both for the overall private sector and for manufacturing and services separately. In addition, we find a
weakly significant positive competition effect for unadjusted productivity growth in the manufacturing
sector. Regarding adjusted employment growth, the competition effect is negative for the private
economy as a whole but insignificant for manufacturing and services separately. Moreover, we find
that population density is positive for adjusted growth in both productivity and employment.
5. CONCLUSIONS
In this paper we tested the role that related variety in education and occupation play in regional
growth. While most analyses of regional variety put firms and sectors in center stage, we argue that
knowledge spillovers occur primarily between individuals. Relatedness between individuals in terms
of education and occupation may be at least as important as industries in stimulating knowledge
spillovers and regional growth. To test the empirical relevance of these arguments we added measures
of education and occupational variety (related and unrelated) to the basic empirical model of Frenken
et al. (2007) and estimated the relative importance of the three dimensions of related variety. As in
previous findings, we found that related variety in industries has a positive effect on growth in
employment but a negative effect on growth in productivity.
However, we also show that related variety in education is positively associated with productivity
growth in the private sector in general and the manufacturing sector in particular, while related variety
in occupation is positively related to productivity growth in the service sector. These results broadly
support that relatedness in terms of the education and occupation of employees are conducive for
knowledge spillovers that stimulate the productivity growth of a region. The potential of productive
interactions between employees in a region is greater when there is related variety in their knowledge
base.
18
The findings also bear on policy in the sense that they illustrate the multi-dimensional nature of the
notion of relatedness. The recent policy concept of smart specialization suggest for instance that
regions should focus on locally strong areas, and develop into areas that are related to these. The way
in which relatedness is conceptualized clearly plays a crucial role in such efforts. Relatedness in
industries and relatedness in employee education and skills do not necessarily go hand-in-hand,
especially in view of that many regions are functionally specialized such that the functions within a
given industry differ widely across regions.
While the results in this paper is consistent with the idea that knowledge spillovers works better in
contexts when related variety is high, it should be stated that the analysis does not inform about the
mechanisms. For example, we do not know whether the estimated relationships reflect pure
knowledge spillovers, better local matching efficiency, or embodied knowledge flows mediated by
local inter-firm labor mobility. Related variety in either industries, education or occupation could
stimulate either of these mechanism. Further work may focus on untangling the mechanism behind
the empirical regularity of a strong association between related variety and growth.
REFERENCES Aiginger, K. and Davies, S. W. (2004). "Industrial specialization and geographic concentration: Two
sides of the same coin? Not for the European Union." Journal of Applied Economics 7(2): 231-248.
Almeida, P. and Kogut, B. (1999). "Localization of knowledge and the mobility of engineers in
regional networks." Management Science 45(7): 905-917.
Andersson, M., Klaesson, J. and Larsson, J. P. (2012). How local are spatial density externalities?
Evidence from square grid data. CIRCLE Paper No. 2012/10.
Andersson, M. and Lööf, H. (2011). "Agglomeration and productivity: Evidence from firm-level
data." The Annals of Regional Science 46(3): 601-620.
Anselin, L. (2003). Spatial econometrics. A companion to theoretical econometrics. Baltagi, B. H.
Oxford, Wiley-Blackwell: 310-330.
Attaran, M. (1986). "Industrial diversity and economic performance in U.S. areas." The Annals of
Regional Science 20(2): 44-54.
Autor, D. H., Levy, F. and Murnane, R. J. (2003). "The skill content of recent technological change:
An empirical exploration." The Quarterly Journal of Economics 118(4): 1279-1333.
Bacolod, M., Blum, B. S. and Strange, W. C. (2009). "Skills in the city." Journal of Urban Economics
65(2): 136-153.
19
Bishop, P. and Gripaios, P. (2010). "Spatial externalities, relatedness and sector employment growth in
Great Britain." Regional Studies 44(4): 443-454.
Boschma, R. (2005). "Proximity and innovation: a critical assessment." Regional Studies 39(1): 61-74.
Boschma, R. and Iammarino, S. (2009). "Related variety, trade linkages, and regional growth in Italy."
Economic Geography 85(3): 289-311.
Boschma, R. and Martin, R. (2010). The aims and scope of evolutionary economic geography. The
handbook of evolutionary economic geography. Boschma, R. and Martin, R. Cheltenham, Edward
Elgar.
Boschma, R., Minondo, A. and Navarro, M. (2012). "Related variety and regional growth in Spain."
Papers in Regional Science 91(2): 241-256.
Brachert, M., Kubis, A. and Titze, M. (2011). Related variety, unrelated variety and regional
functions: Identifying sources of regional employment growth in Germany from 2003 to 2008. IWH-
Diskussionspapiere No. 2011(15).
Cohen, W. M. and Levinthal, D. A. (1990). "Absorptive capacity: a new perspective on learning and
innovation." Administrative Science Quarterly 35(1): 128-152.
Desrochers, P. and Leppälä, S. (2011). "Opening up the ‘Jacobs Spillovers’ black box: local diversity,
creativity and the processes underlying new combinations." Journal of Economic Geography 11(5):
843-863.
Duranton, G. and Puga, D. (2000). "Diversity and Specialisation in Cities: Why, where and when does
it matter?" Urban Studies 37(3): 533-555.
Duranton, G. and Puga, D. (2004). Micro-foundations of urban agglomeration economies. Handbook
of Regional and Urban Economics. Henderson, J. V. and Jacques-François, T., Elsevier. Volume 4:
2063-2117.
Duranton, G. and Puga, D. (2005). "From sectoral to functional urban specialisation." Journal of
Urban Economics 57(2): 343-370.
Easterly, W. and Levine, R. (2002). It's not factor accumulation: Stylized facts and growth models.
Economic growth: Sources, trends and cycles. Loayza, N. and Soto, R., Central Bank of Chile. 6: 61-
114.
Ejermo, O. (2005). "Technological diversity and Jacobs’ externality hypothesis revisited." Growth and
Change 36(2): 167-195.
Essletzbichler, J. (2007). Diversity, stability and regional growth in the United States, 1975-2002.
Applied Evolutionary Economics and Economic Geography. Frenken, K. Cheltenham, Edward Elgar.
Frenken, K., Oort, F. G. V., Verburg, T. and Boschma, R. A. (2005). Variety and regional economic
growth in the Netherlands. Space and Productivity, Ministry of Economic Affairs, The Netherlands.
Frenken, K., Van Oort, F. and Verburg, T. (2007). "Related variety, unrelated variety and regional
economic growth." Regional Studies 41(5): 685-697.
Gibbons, S. and Overman, H. G. (2012). "Mostly pointless spatial econometrics?" Journal of Regional
Science 52(2): 172-191.
20
Glaeser, E. L., Kallal, H. D., Scheinkman, J. A. and Shleifer, A. (1992). "Growth in cities." Journal of
Political Economy 100(6): 1126-1152.
Glaeser, E. L. and Scheinkman, J. A. (2003). Non-market interactions. Advances in economics and
econometrics: Theory and applications. Dewatripont, M., Hansen, L. P. and Turnovsky, S. J.
Cambridge, Cambridge University Press. 3: 339-370.
Hall, B. H. (2011) Using productivity growth as an innovation indicator. Report for the High Level on
Measuring Innovation, DG Research, European Commision
Hall, B. H. and Mairesse, J. (2006). "Empirical studies of innovation in the knowledge-driven
economy." Economics of Innovation and New Technology 15(4-5): 289-299.
Hartog, M., Boschma, R. and Sotarauta, M. (2012). "The impact of related variety on regional
employment growth in Finland 1993–2006: High-tech versus medium/low-Tech." Industry and
Innovation 19(6): 459-476.
Henderson, V., Kuncoro, A. and Turner, M. (1995). "Industrial development in cities." Journal of
Political Economy 103(5): 1067-1090.
Jacquemin, A. P. and Berry, C. H. (1979). "Entropy measure of diversification and corporate growth."
The Journal of Industrial Economics 27(4): 359-369.
Lucas Jr., R. E. (1988). "On the mechanics of economic development." Journal of Monetary
Economics 22(1): 3-42.
Lööf, H. and Heshmati, A. (2002). "Knowledge capital and performance heterogeneity: A firm-level
innovation study." International Journal of Production Economics 76(1): 61-85.
Maskell, P. and Malmberg, A. (1999). "Localised learning and industrial competitiveness." Cambridge
Journal of Economics 23(2): 167-185.
McCann, P. and Ortega-Argilés, R. (2013). "Smart specialization, regional growth and applications to
European Union cohesion policy." Regional Studies: 1-12.
Neffke, F. and Henning, M. (2013). "Skill relatedness and firm diversification." Strategic Management
Journal 34(3): 297-316.
Noteboom, B. (2000). Learning and innovation in organizations and economies. Oxford, Oxford
University Press.
Palan, N. (2010). Measurement of specialization - The choice of indices. FIW Working Paper No 62.
Porter, M. (2003). "The economic performance of regions." Regional Studies 37(6-7): 545-546.
Power, D. and Lundmark, M. (2004). "Working through knowledge pools: labour market dynamics,
the transference of knowledge and ideas, and industrial clusters." Urban Studies 41(5/6): 1025-1044.
Rauch, J. E. (1993). "Productivity gains from geographic concentration of human capital: Evidence
from the cities." Journal of Urban Economics 34(3): 380-400.
Rosenthal, S. S. and Strange, W. C. (2008). "The attenuation of human capital spillovers." Journal of
Urban Economics 64(2): 373-389.
21
Saviotti, P. and Frenken, K. (2008). "Export variety and the economic performance of countries."
Journal of Evolutionary Economics 18(2): 201-218.
Van Oort, F. (2013). Unity in variety? Agglomeration economics beyond the specialization-diversity
controversy. Papers in Evolutionary Economic Geography, Utrecht University.
Van Stel, A. and Storey, D. (2004). "The Link between Firm Births and Job Creation: Is there a Upas
Tree Effect?" Regional Studies 38(8): 893-909.
22
APPENDIX 1
Table A1. Educational codes.
Educational code
(Sun2000Inr)
Education focus
01 General education
08 Reading and writing for adults
09 Personal development
14 Pedagogics and teaching
21 Arts and media
22 The humanities
31 Social and behavioral science
32 Journalism and information
34 Business
38 Law and legal science
42 Biology and environmental science
44 Physics, chemistry and geoscience
46 Mathematics and natural science
48 Computer science
52 Engineering: Technical, mechanical, chemical and electronics
54 Engineering: Manufacturing
58 Engineering: Construction
62 Agriculture
64 Animal healthcare
72 Healthcare
76 Social work
81 Personal services
84 Transport services
85 Environmental care
86 Security
99 Unknown
Table A2. Occupational codes.
ISCO/SSYK
code
Occupation
0 Militaries
1 Managers, legislators and senior officials
21 & 31 Physical, mathematical and engineering science professionals and associate
professionals
22 & 32 Life science and health professionals and associate professionals
23 & 33 Teaching professionals and associate professionals
24 & 34 Other professionals and associate professionals
4 Office and customer services clerks
5 Salespersons, demonstrators, personal and protective services workers
6 Market-oriented skilled agricultural and fishery workers
7 Extraction, building, metal, machinery, handicraft and related trades workers
81 Stationary-plant, machine, mobile-plant and related operators
91 Sales and services elementary occupations, agricultural, mining, transport
and related laborers
23
APPENDIX 2
RV Industry RV Education RV Occupation
UV Industry UV Education UV Occupation
Figure A1. Quantile maps of related and unrelated variety, the private sector (darker color implies
greater variety).
24
RV Industry RV Education RV Occupation
UV Industry UV Education UV Occupation
Figure A2. Quantile maps of related and unrelated variety, the manufacturing sector (darker color
implies greater variety).
25
RV Industry RV Education RV Occupation
UV Industry UV Education UV Occupation
Figure A3. Quantile maps of related and unrelated variety, the service sector (darker color implies
greater variety).
26
Figure A4. Quantile map of the population density in Sweden 2002 (darker color implies greater
population density).
Stockholm Gothenburg
Malmö
27
APPENDIX 3 Table A3. Descriptive statistics for the private sector.
Mean Std. dev. Min Max
Productivity growth 1.26 .178 .824 2.28
Productivity growth adjusted 1.14 .412 .519 5.70
Employment growth 1.06 .091 .726 1.45
Employment growth adjusted 1.03 .165 .539 1.82
RV Industry 1.33 .344 .455 2.16
UV Industry 2.68 .234 1.69 3.12
RV Education 1.52 .138 1.18 1.86
UV Education 2.14 .177 1.69 2.88
RV Occupation 1.55 .129 .885 1.82
UV Occupation 1.97 .118 1.58 2.25
Competition .142 .050 .051 .303
Pop. density 125 420 .257 4,040
Capital-labor growth 1.18 .422 .353 4.57
Productivity 02 521 107 250 1,070
Employment 02 8,430 24,500 285 353,000
Table A4. Descriptive statistics for the manufacturing sector.
Mean Std. dev. Min Max
Productivity growth 1.30 .240 .711 2.84
Productivity growth adjusted 1.01 .388 .432 4.90
Employment growth 1.03 .128 .656 1.71
Employment growth adjusted .923 .266 .357 2.66
RV Industry .535 .151 .102 1.13
UV Industry 1.32 .260 .420 2.18
RV Education 1.43 .157 .948 1.83
UV Education 2.10 .177 1.69 2.88
RV Occupation 1.53 .190 .573 1.86
UV Occupation 1.69 .144 1.20 2.00
Competition .102 .057 .018 .391
Capital-labor growth 1.25 .576 .299 4.59
Productivity 02 545 153 196 1,350
Employment 02 3,480 6,120 101 67,700
Table A5. Descriptive statistics for the service sector.
Mean Std. dev. Min Max
Productivity growth 1.22 .214 .861 2.86
Productivity growth adjusted 1.25 .696 .566 9.97
Employment growth 1.10 .117 .737 1.56
Employment growth adjusted 1.10 .201 .667 2.52
RV Industry .791 .331 .241 1.78
UV Industry 1.37 .292 .627 2.09
RV Education 1.45 .175 1.04 1.85
UV Education 2.12 .187 1.67 2.88
RV Occupation 1.23 .137 .855 1.54
UV Occupation 2.04 .071 1.61 2.29
Competition .200 .051 .050 .317
Capital-labor growth 1.15 .577 .257 6.83
Productivity 02 475 63.4 333 823
Employment 02 4,950 18,900 181 285,000
28
APPENDIX 4 Table A6. Correlation matrix for independent variables (in logarithmic form), the private sector.
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
1. RV Industry 1
2. UV Industry .620 1
3. RV Education .488 .261 1
4. UV Education .571 .331 .664 1
5. RV Occupation .376 .574 .120 .002 1
6. UV Occupation .693 .577 .569 .668 .150 1
7. Competition .256 .229 -.276 .158 -.162 .371 1
8. Pop. density .384 .062 .449 .425 .007 .273 -.167 1
9. Capital-labor gr. -.108 -.105 -.139 -.119 -.055 -.113 .020 -.097 1
10. Productivity 02 .004 -.120 .414 .156 .038 .037 -.441 .389 -.248 1
11. Employment 02 .582 .348 .693 .471 .372 .326 -.512 .614 -.118 .412 1
Table A7. Correlation matrix for independent variables (in logarithmic form), the manufacturing
sector.
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
1. RV Industry 1
2. UV Industry .525 1
3. RV Education -.073 -.067 1
4. UV Education -.240 -.474 .606 1
5. RV Occupation .407 .250 .272 .248 1
6. UV Occupation .015 -.141 .487 .602 .379 1
7. Competition .069 -.374 -.478 .099 .048 .159 1
8. Pop. density -.167 -.307 .464 .511 .108 .345 -.105 1
9. Capital-labor gr. -.032 -.044 -.055 -.093 -.082 -.157 -.023 -.086 1
10. Productivity 02 -.097 -.010 .452 .217 .022 .169 -.391 .315 -.183 1
11. Employment 02 .123 .099 .794 .467 .372 .332 -.572 .552 -.023 .409 1
Table A8. Correlation matrix for independent variables (in logarithmic form), the service sector.
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
1. RV Industry 1
2. UV Industry .897 1
3. RV Education .750 .636 1
4. UV Education .786 .791 .725 1
5. RV Occupation .568 .409 .650 .466 1
6. UV Occupation .230 .173 .200 .189 .366 1
7. Competition -.403 -.358 -.638 -.436 -.523 .143 1
8. Pop. density .456 .267 .456 .433 .496 .099 -.323 1
9. Capital-labor gr. -.053 -.033 -.183 -.063 -.141 -.086 .095 -.042 1
10. Productivity 02 .346 .205 .498 .375 .469 .092 -.476 .553 -.253 1
11. Employment 02 .752 .586 .817 .655 .745 .149 -.743 .610 -.138 .564 1
29
APPENDIX 5
In the spatial error model the error term of equation 5, εr, is replaced by ur where:
in which Wr is a spatial weight matrix taking the travel time distances between municipalities into
account. λ is the spatial error coefficient, which is equal to zero when there is no spatial correlation in
the error terms. The spatial lag model includes a spatially lagged dependent variable as an additional
explanatory variable. Hence, is introduced on the right hand side of Equation 5. ρ is the
spatial coefficient, which is zero when there is no spatial dependence, implying that the municipal
growth is independent of the growth in neighboring municipalities. (Anselin 2003) Model selection is
based on maximum likelihood values, which implies that in each case the model with the largest
maximum likelihood value is presented.
Table A9. Estimated coefficients for the spatial models regarding productivity growth and
employment growth. Standard errors in brackets.
Variables 2b’
Private
Employment
ML(lag)
Adjusted
3b’
Manufacturing
Productivity
ML(error)
Adjusted
4a’
Services
Productivity
ML(error)
6a’
Services
Employment
ML(error)
6b’
Services
Employment
ML (lag)
Adjusted
RV Industry 0.157***
(0.050)
0.081
(0.058)
-0.133***
(0.030)
0.074***
(0.024)
0.043
(0.040)
UV Industry 0.049
(0.139)
0.159
(0.101)
RV Education 0.050
(0.169)
1.35***
(0.251)
0.000
(0.123)
0.192**
(0.098)
0.194
(0.153)
UV Education -0.298*
(0.163)
0.260
(0.309)
0.726***
(0.147)
-0.246**
(0.123)
-0.170
(0.191)
RV Occupation 0.005
(0.129)
0.068
(0.123)
0.247**
(0.099)
-0.029
(0.081)
0.071
(0.131)
UV Occupation 0.306
(0.255)
0.051
(0.223)
-0.410*
(0.242)
-0.510***
(0.196)
-0.370
(0.322)
Competition -0.094***
(0.036)
0.048
(0.047)
-0.045
(0.038)
0.093***
(0.030)
0.016
(0.050)
Population density 0.016**
(0.007)
0.028**
(0.014)
0.010
(0.007)
0.011*
(0.006)
0.010
(0.007)
Capital labor growth -0.014
(0.036)
0.063***
(0.019)
Productivity 2002 -0.161**
(0.067)
-0.503***
(0.076)
Constant -0.326*
(0.194)
0.297
(0.437)
2.83***
(0.478)
0.713***
(0.169)
0.366
(0.280)
Spatial coefficient 0.333**
(0.145)
0.365**
(0.154)
0.260
(0.177)
0.478***
(0.143)
0.338**
(0.156)
R2 0.236 0.346 0.255 0.128 0.088
Observations 289 289 289 289 289
Notes: *** p<0.01, ** p<0.05, * p<0.1. R2 is a psudo-R
2 corresponding to the variance ratio.
